Discipline: Computer Sciences and Information Management
Subcategory: STEM Science and Mathematics Education
Gary Holness - Optical Science Center for Applied Research, Delaware State University
Co-Author(s): David Pokrajac, Tomasz Smolinski, Sokratis Makrogiannis, and Jinjie Liu, Optical Science Center for Applied Research, Delaware State University
Spectral methods based on laser scattering and spin polarization present new challenges to the identification of bio-macro molecules. Our contributions to recognition of pattern phenomena from bio molecules from their spectra include a model for optimal classifier design based on statistical detection theory, simulations of nonlinear Maxwell scattering using Finite Difference Time Domain (FDTD) methods, non-linear heat diffusion equation models for spatio-temporal cell segmentation from time lapse image sequences, a new information theoretic method called Chisini Jensen Shannon divergences and a new kernel that achieves superior results in Support Vector Machine classification. For cell segmentation, moving regions are initially detected in each set of three consecutive sequence images by numerically solving a system of coupled spatio-temporal partial differential equations and determining the optimal values for the temporal and spatial diffusion parameters. After the spatio-temporal diffusion stage is completed, we compute the edge map. This process is followed by watershed-based segmentation to detect the moving cells. We applied this method on several datasets of fluorescence microscopy images. We validated all segmentation techniques against reference masks provided by the international Cell Tracking Challenge consortium. Our proposed method produced encouraging segmentation accuracy, especially when applied to images containing cells undergoing mitosis and low SNR. The proposed technique yielded average improvements of 7% in segmentation accuracy compared to both strictly spatial and temporally linked Chan-Vese techniques. Using the apparatus of statistical theory of detection, we develop the optimal classifier for spectroscopy data for a linear model of an echelle spectrograph system. We validate model assumptions through statistical analysis of ‘dark signal’ and laser-breakdown induced spectra of standardized NIST glass. The experimental results suggest that the quadratic classifier may provide optimal performance if the spectroscopy signal and noise can be considered Gaussian. We perform multi-class classification of laser-Induced breakdown spectroscopy data for four commercial samples of proteins: bovine serum albumin, osteopontin, leptin, and insulin-like growth factor II. We achieve classification accuracies above 98% by using the linear classifier with 21-31 principal components.
Funder Acknowledgement(s): This work was supported by National Science Foundation CREST grant - HRD-1242067.
Faculty Advisor: None Listed,